Electrical Tools Object Detection

📘 Project Overview

This project focuses on object detection of electrical tools using YOLOv8 and a custom-labeled dataset created with Roboflow. The goal was to accurately detect and classify different electrical tools such as drills, hammers, screwdrivers, and pliers from real-world images. A total of 500 images were collected from Google Images using Chrome WebDriver, labeled in Roboflow, and split into training, validation, and testing sets. The model was trained, evaluated, and finally deployed using a Streamlit web app for real-time detection on images and videos.

---

Data

Due to large file sizes, the full dataset is hosted on Kaggle instead of GitHub.
You can access and download the dataset directly using the following link:

🔗 Kaggle Dataset: Electrical Tools Object Detection Dataset

📁 Dataset Details

• Total Images: 500+
• Classes: ['brush', 'drill machine', 'fine point pliers', 'hammer', 'knife', 'pliers', 'scissor', 'screwdrivers', 'spanners', 'tape']
• Split Ratio:
  • 🟩 Train: 70%
  • 🟨 Validation: 10%
  • 🟦 Test: 20%
• Format: YOLOv8 (Images + .txt Labels)
• Labeled Using: Roboflow

---

🧩 Project Pipeline

1. Data Collection

• Images of 10 electrical tools were scraped from Google Images using Chrome WebDriver.
• Collected images were stored in a single folder and uploaded to Roboflow.

2. Labeling & Dataset Preparation

• In Roboflow, 500 images were labeled manually with class names.
• Classes:['brush', 'drill machine', 'fine point pliers', 'hammer', 'knife', 'pliers', 'scissor', 'screwdrivers', 'spanners', 'tape']
• Roboflow handled data preprocessing, resizing, and augmentation.
• Dataset split: Train 70%, Validation 10%, Test 20%.
• Exported dataset in YOLOv8 format, which includes: train-img,labels valid-img,labels test-img,labels

---

3. Data Validation

• Verified that each image file has a corresponding label (.txt) file.
• Checked and removed mismatched or irregular files.
• Confirmed dataset integrity before training.

---

4. Model Training

• Pretrained weights YOLOv11n.pt (YOLOv8-compatible) were downloaded from GitHub.
• Model initialized and trained using Ultralytics YOLO framework.

5. Model Evaluation & Testing

• Model evaluated on validation and test sets.

6. Deployment

• The trained model was deployed using Streamlit for real-time image and video inference.
• The app allows users to upload images or use webcam feeds for object detection.

---

🧠 Technologies Used

• Python
• YOLOv8 / YOLOv11n (Ultralytics)
• Roboflow (for data labeling and preprocessing)
• Streamlit (for deployment)
• OpenCV, NumPy, Pandas, Matplotlib

---

🚀 Key Highlights

✅ 500+ images collected and labeled manually. ✅ Model trained using YOLOv8 with Roboflow integration. ✅ Streamlit app for real-time detection on images and videos. ✅ Custom dataset validation and cleaning scripts.

---

📈 Future Improvements

• Increase dataset size for better generalization.
• Fine-tune confidence thresholds and augmentations.
• Deploy app on Render / Hugging Face / Streamlit Cloud.

---

🧾 Acknowledgements

• Roboflow for dataset labeling and preprocessing tools.
• Ultralytics YOLO for an efficient object detection framework.
• Kaggle for training and validation environment.

---

👤 Author

Venktesh Deep Learning & Computer Vision Enthusiast 📧 [www.linkedin.com/in/venkatesh-ds25] 🔗 [https://github.com/VK-Venkatesh]

---

“Teaching machines to see the world of electrical tools – one frame at a time.”